4. Manage data with the AML Dashboard

4.1. Background

The AML Dashboard is a web-based tool that allows users to interact with the AML framework.

This tutorial showcases the data collection step of an AML model training process using the AML Dashboard.

4.2. Prerequisites

Ensure you have installed the AML Dashboard using one of the following methods:

For more information, check the AML Dashboard Interfaces and AML Dashboard Usage sections.

4.3. Running the demo

To run the necessary components for collecting data using the AML Dashboard, follow these steps:

4.3.1. Start the backend server

  1. Navigate to the backend directory.

cd backend

  1. Load the AML-IP environment.

source /AML-IP/install/setup.bash

  1. Start the server:

python3 server.py

4.3.2. Start the AML Dashboard

  1. Navigate to the frontend/aml_dashboard directory.

cd frontend/aml_dashboard

  1. Start the AML Dashboard:

npm run dev

  1. Access the dashboard at http://localhost:5173/.

4.4. Collecting data

There are several options available for dataset creation:

  • Webcam: Record instances using the webcam.

  1. Navigate to the Data Management tab on the AML Dashboard.

  2. Choose Sensors from the drop-down menu in the Choose the model for the training set section.

  3. In the webcam section, toggle the activate video button to enable the webcam.

  4. Introduce a label for the class that will be recorded in the Instance label section.

  5. Press the Hold to record instances button in the Capture instances to the training set section to start recording data.

  • Standard dataset: Load a standard dataset from the system.

  1. Navigate to the Data Management tab on the AML Dashboard.

  2. Choose a dataset from the drop-down menu in the Choose the model for the training set section.

  3. Click on the Load dataset button to load the dataset.

  • Custom dataset: Load a custom dataset.

  1. Navigate to the Data Management tab on the AML Dashboard.

  2. Choose ‘Custom’ from the drop-down menu in the Choose the model for the training set section.

  3. Click on the Load dataset button to load the custom dataset.

  4. A popup will appear, allowing you to select the desired dataset from your local machine.

Warning

The custom dataset must be a json file in the correct format.

To create the dataset in the desired format, the code snippet below can be used as a reference:

def save_in_format(images, labels):
   """
   Function that formats the data to be saved in the required format.
   Arguments:
   images: List of images. Each image is a one-dimensional list.
   labels: List of labels. Each label is an integer.
   returns: Formatted data
   """

   def generate_data_uri(image):
       """
       Convert a 2D image list or array to a PNG data URI.
       """
       # Convert the list to a NumPy array for processing
       image_len = len(image)
       one_side = int(math.sqrt(image_len))
       image_array = np.array(image, dtype=np.uint8).reshape(one_side, one_side)
       img = Image.fromarray(image_array, mode="L")  # Create a PIL image from the array
       buffered = io.BytesIO()
       img.save(buffered, format="PNG")  # Save the image to the buffer in PNG format
       img_data = buffered.getvalue()
       data_uri = f"data:image/png;base64,{base64.b64encode(img_data).decode('utf-8')}"
       return data_uri

   combined_data = list(zip(images, labels))
   sampled_data = random.sample(combined_data, 100)
   sampled_images, sampled_labels = zip(*sampled_data)
   # Prepare the data in the required format
   formatted_data = {
       "total": len(images),
       "limit": 1000,
       "skip": 0,
       "instances": [
            {
                "datasetName": "training2-set-models", # This must be kept as is
                "x": image, # The image must be a one-dimensional list
                "thumbnail": generate_data_uri(image),
                "y": label, # The label must be numeric
                "id": idx,
            }
            for idx, (image, label) in enumerate(zip(sampled_images, sampled_labels))
         ]
      }

   return formatted_data